Motor-acoustic mappings shape child phonology:Evidence from a circular chain shift

Tara McAllister Byun1 Adam Buchwald1

1New York University, Department of Communicative Sciences and Disorders

1 / 25

Child chain shifts

I A topic of perennial interest in the child phonology literature:chain shifts that appear to arise spontaneously indevelopment (e.g., Dinnsen & Barlow, 1998; Jesney, 2007; Rose, 2009,

Ettlinger, 2009; Dinnsen et al., 2011)

I Chain shift: Interacting phonological processes causesuccessive changes along some dimension (A → B; B → C)

I e.g. sun → [T2n], thumb → [f2m] (Dinnsen & Barlow, 1998)

I puzzle → [p2d@l], puddle → [p2g@l] (Smith, 1973; Jesney,2007; Dinnsen et al., 2011)

2 / 25

Accounts of child chain shifts

I A case of opacity, i.e. phonological generalizations that arenot surface-true.

I Problematic for constraint-based grammars:I Why doesn’t the constraint that drives labialization in

thumb → [f2m] also apply in sun → [T2n]?

I Performance limitations do not appear to offer a solution:I If child is capable of producing [T], why does he/she not deploy

it in the intended context?

I This talk will make the case that a grammar that incorporatesperformance pressures (motor, perceptual) can capture evenhighly problematic cases of chain shift.

I A-map model (McAllister Byun, Inkelas, & Rose, in press)I Linked Attractor model (Menn, Schmidt, & Nicholas, 2009)

3 / 25

Circular chain shift?

I We present an apparent case of circular chain shift (A → B;B → A) in a child with minor phonological delay.

I Initial homorganic s-stop clusters were reduced (stick → [sIk])

I But at the same time, [t] epenthesis converted initial coronalsingletons to clusters (sick → [stIk]).

I Not an easy phenomenon to capture in a formal grammar:I “The existence of a circular chain shift in which all links occur

synchronically would present a problem for the OT doctrine ofharmonic ascent...Moreton (1999) provides a formal proofshowing that an OT grammar that admits only faithfulnessand markedness constraints is incapable of modeling circularchain shifts” (Crowhurst, 2011)

I But hardly a straightforward performance phenomenon, either!

4 / 25

Case study

I “Wesley,” initially evaluated age 3;7

I Strong expressive and receptive language abilities

I History of mild speech delay

I Score on Hodson Assessment of Phonological Patterns-3 fell1.25 SD below mean for age

I Decreased intelligibility due to multiple phonological patterns.

5 / 25

Case study

I Velar fronting (all positions), palatal fronting (inconsistent)

I got to chew gum

I Reduction of /s/-obstruent clusters in initial positionI Affected /st/, /sk/ clusters:

I stop and go

I in the sky

I But not /sp/ clusters:I a spoon

6 / 25

Exploratory therapy sessions

I Multiple oppositions (Williams, 1993, 2000, 2003) approachtargeting /s/-/st/-/sk/ contrasts

I Session 1: Initially unable to imitate clusters.I store

I Later in Session 1: Able to produce clusters with cueing,inconsistently.

I Good SLP!

I But also started to insert stops in singleton fricative contexts.

I Bad SLP!

7 / 25

Exploratory therapy sessions

I Session 2: More accurate cluster productions, but more stopinsertion with singleton targets as well.

Session 1 Session 2

0

5

10

15

20

st s st sTarget

coun

t OutputAccurateDeletionEpenthesis

Figure 1: Realization of cluster and singleton targets across treatmentsessions

8 / 25

Exploratory therapy sessionsI Session 3: Produces clusters for singletons more often than for

cluster targets.I Session 4: Finally starting to resolve overgeneralization.

Session 3 Session 4

0

5

10

15

20

25

st s st sTarget

coun

t OutputAccurateDeletionEpenthesis

Figure 2: Realization of cluster and singleton targets across treatmentsessions

9 / 25

A perceptual or representational problem?

I Perceptual testing: Forced-choice picture-pointing task inresponse to experimenter’s verbal model.

I 10/10 correct responses for a /st/-/s/ minimal pairI 10/10 correct responses for a /st/-/sk/ minimal pair

I Wesley does perceive the /s/-/st/-/sk/ contrast.

I And he can map the perceived contrast to distinct storedlexical representations.

10 / 25

Covert contrast?I Is the [st] that Wesley produces in error for target /s/

identical to the [st] that he produces for target /st/?I Is the [s] that he produces for target /s/ identical to the [s]

that he produces for target /st/?I We measured minimal pairs sick-stick and sir-stir to look for

covert contrast in Wesley’s output (Table 1).

Target Realized with [s] Realized with [st]

sir 4 10stir 2 12sick 7 23stick 7 12

Table 1: Count of tokens realized with cluster versus singleton

I Measures included closure duration (surface [st] only), VOT(surface [st] only), fricative duration (all tokens).

11 / 25

Covert contrast?I Underlying clusters (/st/ → [st]) did not differ significantly

from derived clusters (/s/ → [st])I With respect to closure durationI With respect to VOT

0

200

400

s stTarget

Mea

n cl

osur

e du

ratio

n (m

s)

Figure 3: Closure duration in trueversus derived clusters

0

10

20

30

s stTarget

VO

T (

ms)

Figure 4: VOT in true versusderived clusters

12 / 25

Covert contrast?I Similarly, no difference in fricative duration:

I between underlying and derived singletons (/s/ → [s] versus/st/ → [s])

I between underlying and derived clusters (/st/ → [st] versus/s/ → [st]).

I In short, no evidence of covert contrast.

0

100

200

300

400

500

Singleton ClusterOutput

Mea

n fr

icat

ive

dura

tion

(ms)

Target_Codesst

Figure 5: Differences in fricative duration

13 / 25

Covert contrast?I There was a significant difference in fricative duration between

surface clusters and singletons, independent of underlying orderived status (t =4.4, df = 78, p < .0001).

I Contrary to expectations from adult speech, [s] in surfacecluster contexts was significantly longer than singleton [s].

0

100

200

300

400

500

Singleton ClusterOutput

Mea

n fr

icat

ive

dura

tion

(ms)

Target_Codesst

Figure 6: Differences in fricative duration

14 / 25

Gestural coordinationI Wesley’s earliest [st] productions featured less than typical

degree of coarticulatory overlap.I Suggestive of difficulty with gestural coordination of multiple

consonants (Davidson, 2006; Miozzo & Buchwald, 2012).I Resembles gestural mistiming described in adult production of

non-native clusters.

Figure 7: An articulatory-driven repair of a non-native consonant clustersequence (image from Davidson, 2006)

I In homorganic clusters, sequence of slightly differentmovements of a single articulator may represent a particulararticulatory challenge (Bates, Watson, & Scobbie, 2002).

15 / 25

An articulatory pressure, a phonological repair

I If articulatory difficulty is the driving force behind Wesley’scluster reduction, should we analyze these outputs asextragrammatical performance errors (Hale & Reiss 1998, 2008)?

I No, there is a specific reason to reject this analysis:I In a C1C2V cluster, there is tighter gestural coupling between

C2 and V than C1 and V (Nam, Goldstein, & Saltzman, 2010).I In cases of cluster reduction as articulatory performance error,

expect to observe reduction to C2 (Miozzo & Buchwald, 2012).

16 / 25

An articulatory pressure, a phonological repair

I So why does Wesley produce [sIk] instead of [tIk] for “stick”?

I Because the articulatory pressure interacts with other factors.I Goodness of perceptual match for adult target:

I /s/ is acoustically salientI Reduction of /st/ to [t] is a greater perceptual deviation than

reduction to [s]

I Vacuous coalescence (/st, sk/ → [s]) addresses thearticulatory challenge while achieving closer match for adulttarget.

17 / 25

Influence of motor learning on preferred repair

I What changed to allow the emergence (andovergeneralization) of the output in which both segments ofthe cluster were preserved?

I Our contention: A change in the availability of a stable motorplan.

I In the therapy setting, Wesley identified and stabilized themotor routine for cluster production with with minimalgestural overlap.

I And due to recent practice, sometimes activated in non-targetcontexts.

I Overgeneralization to singleton targets interpreted as aperformance error reflecting high level of activation of clustermotor routine.

18 / 25

Modeling the motor-grammar interface

I Patterns in speech development and disorders can havetransparent origins in phonetic performance factors...

I ...but they also demonstrably interact with perceptual andstructural/representational factors.

I The A-map model (McAllister Byun, Inkelas, & Rose, in press)aims to integrate performance pressures into the feature-basedformalism that has been so successful in describingpatterns/alternations in fully-developed phonologies.

I The A-map in a nutshell:I Stored knowledge about the reliability of different

motor-acoustic mappings.I Grammatical constraint favors candidates linked to a reliable

plan.I Dynamically updated; gain or loss of motor skill can be

expressed within the grammar.

19 / 25

AssumptionsI Phonetic experience (inputs perceived, outputs produced)

stored as episodic traces in multi-dimensionalauditory-acoustic space.

I Phonological representations linked to phonetic detail (cloudsof traces) via distinctive features in the analysis-by-synthesisframework (Halle & Stevens 1959, 1962; Poeppel, Idsardi &

Wassenhove 2008; Kuhl et al. 2014).20 / 25

AssumptionsI Motor plan executions generate a predicted outcome

(efference copy) in addition to a perceptually encoded output.I For speaker’s own output, both perceptually encoded trace

and trace of efference copy are stored.

Figure 8: Clouds in speaker-transformed auditory-acoustic spacerepresenting the adult target (T), the child’s actual outputs (A), andefference copies representing the expected sensory consequences ofplanned outputs (E).

21 / 25

Metric of goodness of mappingsI When there is an error in motor planning or execution, there is

divergence between perceptually encoded trace and trace ofefference copy.

I For novel/complex motor plan, frequent errors yield largermean distance between predicted and actual acousticconsequences.

I Indexed in the A-map.

22 / 25

Getting it into the grammarI Precise: Penalize a candidate in proportion to the average

distance between pairs of efference copies and actual outputsin the associated motor-acoustic mapping.

I Pressure favoring articulatory reliability exerted by Precisecan come into competition with faithfulness to adult target.

Figure 9:

23 / 25

Conclusions

I Child phonology offers abundant evidence for links betweenmotor and phonological development.

I An exemplar-based grammar that tracks motor-acousticmappings:

I Provides a direct mechanism to capture articulatory andperceptual pressures without abandoning the benefits offormalism;

I Improves our ability to account for formally problematicphenomena like chain shift.

24 / 25

Thanks!

Any questions?

tara.byun@nyu.edubuchwald@nyu.edu

Thanks to NIDCD (NIH R03DC012883).

25 / 25